An artificial intelligence-based eco-friendly column breakwater construction method and system

Abstract

The invention discloses an eco-friendly column group breakwater construction method and system based on artificial intelligence, in order to solve the technical problems of high cost and difficulty in breakwater construction and protect the ecological balance of water quality in the port area; it is urgently necessary to propose a scheme that can judge the model prediction results, evaluate the engineering value and economic benefits, reduce the construction difficulty and preparation time, save costs and improve the effect of wave reduction; this scheme uses wave-related data and machine learning technology to establish a prediction model; and the prediction model is trained to obtain a trained model, and at the same time, the model prediction results are obtained; the optimal column group relative arrangement distance in the model prediction results is searched according to the particle swarm optimization algorithm, and the optimal solution is continuously approached through the iterative optimization process to obtain the final model prediction result of the optimal column group relative arrangement distance.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an artificial intelligence-based eco-friendly column group breakwater construction method and system. Background Art

[0002] As an important component of artificial protection of coastal ports, breakwaters can block the impact of waves, enclose harbors, and maintain a stable water surface to protect ports from bad weather. At the same time, breakwaters can also provide security for ports, ensure the stability of the port's water level, and meet the needs of ship berthing, loading and unloading operations, and entry and exit navigation. The existing methods mainly determine the relative arrangement distance of the breakwater column group through wave simulation analysis and structural stability analysis, including the use of physical models or numerical simulation software to simulate the propagation, reflection, and diffraction of waves in front of the breakwater, and stability analysis of the column group structure. By calculating the stress conditions of the column group at different arrangement distances, the optimal arrangement scheme is determined to ensure the stability and safety of the structure.

[0003] However, the propagation, reflection and diffraction process of waves in front of the breakwater is a complex physical phenomenon, which is affected by many factors. The accuracy and resolution of the existing model are difficult to capture all the key physical phenomena, resulting in differences between the simulation results and the actual situation; and when conducting stability analysis on the column group structure, it is necessary to consider the effects of various external loads. The calculation of these loads is itself a complex problem, which requires accurate wave data and structural parameters. These factors may lead to inaccurate analysis results; therefore, there are still problems with the cost and difficulty of breakwater construction. It is difficult to reduce the construction difficulty and preparation time, and it is impossible to improve the wave reduction effect of the column group breakwater while saving costs. Summary of the invention

[0004] The present invention provides an artificial intelligence-based eco-friendly column group breakwater construction method and system to solve the problem of difficulty in determining the optimal relative arrangement distance of the column groups and to overcome the poor wave reduction effect of the breakwater project.

[0005] The wave-related data of the target sea area is input into the prediction model for prediction, and the model prediction results are optimized based on the particle swarm optimization algorithm to obtain the final model prediction results of the optimal relative arrangement distance of the column group; wherein the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of the wave on the column group and maximizing the stability coefficient of the column group;

[0006] Among them, the prediction model is established by learning and training the initial prediction model according to the feature set of wave parameter data; the feature set is established by fusing the physical derivative features of the wave parameter data with the original features; the initial prediction model is based on the wave parameter data of the target sea area and machine learning technology, and is established by fusing the weights of several parameter features through an inner loop, and based on the knowledge transfer metric value in the outer loop process.

[0007] The present invention can find potential optimal solutions in the entire solution space through the particle swarm optimization algorithm; at the same time, through the local optimization process, the algorithm can continuously approach the optimal solution and improve the quality of the solution. In addition, the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group. The optimization goal is directly aimed at the key performance indicators of the breakwater project, so that the optimization result is more in line with the actual needs. For the prediction model, by fusing the physical derivative features of the wave parameter data with the original features to construct a feature set, this feature fusion technology can more comprehensively reflect the characteristics and laws of the wave data and improve the prediction accuracy of the model. The inner loop optimizes the model's processing and analysis capabilities for wave data by fusing the weights of several parameter features; this optimization enables the model to more accurately capture the subtle changes in wave data and improve the accuracy of prediction. The knowledge transfer metric in the outer loop process is used to further adjust and optimize the model, which helps the model to better utilize historical data and empirical knowledge, improve its applicability in different sea areas, and thus ensure the accuracy of the model prediction.

[0008] Compared with the prior art, in the inner loop of the present invention, the model can optimize the processing and analysis capabilities of wave data by integrating the weights of several parameter features; in the outer loop, the model can be further adjusted and optimized according to the knowledge transfer metric to adapt to the wave characteristics of different sea areas; and, by utilizing the powerful global search capability of the particle swarm optimization algorithm, it is possible to find potential optimal solutions in the entire solution space, thereby solving the problem of difficulty in determining the optimal relative arrangement distance of the column groups to overcome the poor wave reduction effect of the breakwater project.

[0009] As a preferred solution, the initial prediction model is based on the wave parameter data of the target sea area and machine learning technology, and is established by fusing the weights of several parameter features through an inner loop and based on the knowledge transfer metric value in the outer loop process, specifically:

[0010] Acquiring the wave parameter data of the target sea area;

[0011] Dividing the wave parameter data into a plurality of subsets according to a time window, and obtaining a meta-learning task set consisting of the plurality of subsets;

[0012] In the meta-learning task set, several internal network models are trained in an inner loop, and several parameter feature weights in the model forward propagation process are fused to obtain several intermediate models; wherein the several internal network models are established based on the wave parameter data according to machine learning technology;

[0013] The meta-learner is trained based on the several intermediate models and parameter indicator sets, and the initial prediction model is outputted by the meta-learner after the training; wherein the parameter indicator set is established based on the query set performance stability index, the inter-task knowledge transfer metric and the meta-learning weight adjustment factor.

[0014] This preferred solution performs inner loop training on the internal network model in the meta-learning task set, which helps the model fully learn the information in the data and improves the generalization ability of the model. The meta-learner can be trained using several intermediate models and a parameter indicator set established based on the query set performance stability index, the inter-task knowledge transfer metric, and the meta-learning weight adjustment factor. This method can comprehensively consider information from multiple aspects, allowing the meta-learner to learn the characteristics of the data more comprehensively and accurately.

[0015] As a preferred solution, several internal network models are trained in an inner loop, and several parameter feature weights in the forward propagation process of the model are fused to obtain several intermediate models, specifically:

[0016] The plurality of internal network models are trained in an inner loop, and in the forward propagation process of the model, the model is controlled to learn the relative importance of different features through a feature fusion weight learning mechanism, and different features are fused; wherein the feature fusion is performed by treating each feature as a node in a graph neural network and representing the relationship between the features through edges for fusion;

[0017] At the output of the model, measure the similarity between the predicted distance distribution and the actual distance distribution to obtain the loss value of the current iteration round;

[0018] Performing gradient calculation based on the loss value of the current iteration round to obtain the gradient of the loss function relative to the model parameters;

[0019] In each iteration of the inner loop training, dynamically adjusting the learning rate based on an adaptive learning rate adjustment formula;

[0020] Parameters of the several internal network models are updated according to the gradient and the adjusted learning rate. If the current iteration rounds of the several internal network models reach a preset number of iterations or the loss value converges below a preset threshold, the several internal network models after training are defined as the several intermediate models.

[0021] This preferred solution treats each feature as a node in the graph neural network and uses edges to represent the relationship between features for fusion. This method can effectively capture the complex associations and interactions between features. This fusion method is more flexible and powerful than traditional linear combinations or simple splicing, and helps to improve the prediction performance of the model. During the forward propagation of the model, the model is controlled to learn the relative importance of different features through the feature fusion weight learning mechanism, so that the model can automatically adjust the weight of each feature, so as to pay more attention to the key features that have an important impact on the prediction results, and further improve the accuracy and robustness of the model. At the output end of the model, the similarity between the predicted distance distribution and the actual distance distribution is measured to obtain the loss value of the current iteration round. This loss value measurement method can intuitively reflect the prediction performance of the model and provide a reliable basis for subsequent gradient calculations and parameter updates.

[0022] As a preferred solution, the adaptive learning rate adjustment formula is specifically:

[0023]

[0024] in, and are the learning rates of the previous iteration and this iteration respectively, is the learning rate decay coefficient, is a preset positive number. and The loss function is in the parameters and parameters The gradient at .

[0025] As a preferred solution, the parameter index set is established based on the query set performance stability index, the inter-task knowledge transfer metric and the meta-learning weight adjustment factor, specifically:

[0026] Based on the sample prediction values ​​of the several intermediate models, the query set performance stability index is calculated;

[0027] By calculating the distribution difference of parameters of different meta-learning tasks in the reproducing kernel Hilbert space, a knowledge transfer metric value between the tasks is obtained;

[0028] According to the difficulty value of each meta-learning task, dynamically adjust the update amplitude of the parameters by introducing the meta-learning weight adjustment factor;

[0029] The parameter indicator set is composed of the query set performance stability indicator, the inter-task knowledge transfer metric and the meta-learning weight adjustment factor.

[0030] In this preferred solution, the query set performance stability index can reflect the generalization ability and robustness of the model on different samples. By calculating the distribution difference of the parameters of different meta-learning tasks in the reproducing kernel Hilbert space to obtain the knowledge transfer metric between tasks, the similarities and differences between tasks can be quantified. By introducing the meta-learning weight adjustment mechanism, the update amplitude of the parameters can be dynamically adjusted according to the difficulty value of each meta-learning task.

[0031] As a preferred solution, the inter-task knowledge transfer metric is specifically:

[0032] ;

[0033]

[0034] in, is the knowledge transfer metric between the tasks, is an intermediate parameter used to measure the difference in the distribution of two data sets. is the source task dataset, is the target task dataset, is the mapping function that maps data to the reproducing kernel Hilbert space, and are the mean vectors of the source task dataset and the target task dataset in the reproducing kernel Hilbert space, is a hyperparameter, and Respectively represent the number of samples in the source task dataset and the target task dataset, represents the reproducing kernel Hilbert space.

[0035] As a preferred solution, the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group, specifically:

[0036] A time window is set, a weight value is calculated according to the wave data in the time window, and the weight value is continuously updated as the time window slides to obtain a wave weight parameter;

[0037] Fuzzy sets are defined for the two objectives of minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group, and a membership function is constructed for each fuzzy set to obtain a membership function set;

[0038] Based on the wave weight parameters and the membership function set, the multi-objective optimization function is constructed.

[0039] This preferred solution can capture the dynamic changes of waves in real time by setting a time window and calculating the weight value based on the wave data in the window; and as the time window slides, the weight value is continuously updated to reflect the latest situation of the wave conditions. Defining fuzzy sets for the two goals of minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group helps to deal with the uncertainty in these goals. A membership function is constructed for each fuzzy set so that the specific value of each goal can be mapped to a membership between 0 and 1. This helps to balance these two potentially conflicting goals during the optimization process.

[0040] As a preferred solution, the model prediction results are optimized based on the particle swarm optimization algorithm to obtain the final model prediction results for the relative arrangement distance of the optimal column groups, specifically:

[0041] Setting a main particle group and an auxiliary particle group according to the prediction results of the model;

[0042] Searching in the master particle swarm according to the particle swarm optimization algorithm to obtain a global optimal solution;

[0043] In the auxiliary particle group, a combination operation is performed on a number of elite particles and a mutation operation is performed on the number of elite particles to obtain an auxiliary global optimal solution; wherein the number of elite particles are obtained by dividing the auxiliary particle group according to the fitness values ​​of the particles;

[0044] The actual optimal solution among the global optimal solution and the auxiliary global optimal solution is defined as the final model prediction result regarding the relative arrangement distance of the optimal column groups.

[0045] This preferred solution achieves effective division of the search space by setting up a main particle group and an auxiliary particle group. The main particle group focuses on global search and can quickly locate a relatively excellent solution space area. The auxiliary particle group further combines and mutates the elite particles, which helps to perform a detailed search in the located solution space area, thereby improving the search efficiency.

[0046] As a preferred solution, the feature set is established by fusing the physical derivative features of the wave parameter data with the original features, specifically:

[0047] The physical derivative features and the original features of the wave parameter data are mapped to a high-dimensional space based on a kernel function, and the physical derivative features and the original features are fused in the high-dimensional space to obtain the feature set.

[0048] This preferred solution fuses the original features and physically derived features in a high-dimensional space. This fusion method can fully utilize the information of the two types of features to generate a more representative feature set, which helps to improve the prediction accuracy of subsequent models.

[0049] The present invention also provides an artificial intelligence-based eco-friendly column group breakwater construction system, including a prediction module;

[0050] The prediction module is used to input the wave-related data of the target sea area into the prediction model for prediction, and optimize the model prediction results based on the particle swarm optimization algorithm to obtain the final model prediction results of the relative arrangement distance of the optimal column group; wherein the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of the wave on the column group and maximizing the stability coefficient of the column group;

[0051] Among them, the prediction model is established by learning and training the initial prediction model according to the feature set of wave parameter data; the feature set is established by fusing the physical derivative features of the wave parameter data with the original features; the initial prediction model is based on the wave parameter data of the target sea area and machine learning technology, and is established by fusing the weights of several parameter features through an inner loop, and based on the knowledge transfer metric value in the outer loop process. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of an artificial intelligence-based eco-friendly column group breakwater construction method provided in an embodiment of the present application;

[0053] Figure 2 It is a structural schematic diagram of an artificial intelligence-based eco-friendly column group breakwater construction system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0055] In the description of the present application, unless otherwise specified, “several” means two or more.

[0056] Embodiment 1:

[0057] See also Figure 1 The embodiment of the present application provides an artificial intelligence-based eco-friendly column group breakwater construction method, including S1, and the specific implementation steps are as follows:

[0058] S1. Input the wave-related data of the target sea area into the prediction model for prediction, and optimize the model prediction results based on the particle swarm optimization algorithm to obtain the final model prediction results of the optimal relative arrangement distance of the column group; wherein the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of the wave on the column group and maximizing the stability coefficient of the column group;

[0059] Among them, the prediction model is established by learning and training the initial prediction model according to the feature set of wave parameter data; the feature set is established by fusing the physical derivative features of the wave parameter data with the original features; the initial prediction model is based on the wave parameter data of the target sea area and machine learning technology, and is established by fusing the weights of several parameter features through an inner loop, as well as based on the knowledge transfer metric in the outer loop process.

[0060] Step S1 of the embodiment of the present application includes S1.1 to S1.6, specifically:

[0061] S1.1. Regularly record wave data in the target sea area through equipment such as buoys, wave measuring vessels or fixed wave measuring stations, so as to obtain wave parameter data of the target sea area, and clean, normalize or standardize the wave parameter data; wherein the wave parameter data includes key characteristic data such as wave height, wave period, wind direction, wind speed and water depth;

[0062] The wave parameter data is divided into several subsets according to the time window, and each subset contains all the wave parameter data within the time window; the divided data subsets are taken as independent tasks to construct a meta-learning task, thereby obtaining a meta-learning task set composed of several meta-learning tasks; wherein the time window can be a fixed length, such as hourly, daily, weekly or monthly.

[0063] S1.2. Based on the wave parameter data, several internal network models are established according to machine learning techniques;

[0064] In the meta-learning task set, several internal network models are trained in an inner loop, and in the forward propagation process of the model, the relative importance of different features learned by the model is controlled through the feature fusion weight learning mechanism, and different features are fused; wherein, feature fusion is performed by treating each feature as a node in the graph neural network and representing the relationship between features through edges, specifically: using the node representation learning ability of the graph neural network, each feature is regarded as a node in the graph, and the relationship between features is represented by edges; then, the aggregation operation of the graph neural network is used to update the node representation, so as to obtain the fused feature representation;

[0065] At the output of the model, measure the similarity between the predicted distance distribution and the actual distance distribution to obtain the loss value of the current iteration round;

[0066] Perform gradient calculation based on the loss value of the current iteration round to obtain the gradient of the loss function relative to the model parameters;

[0067] In each iteration of the inner loop training, the learning rate is dynamically adjusted based on the adaptive learning rate adjustment formula;

[0068] Parameters of the several internal network models are updated according to the gradient and the adjusted learning rate. If the current iteration rounds of the several internal network models reach a preset number of iterations or the loss value converges below a preset threshold, the trained several internal network models are defined as several intermediate models.

[0069] Among them, the formula of the feature fusion weight learning mechanism is:

[0070]

[0071] The formula for measuring the similarity between the predicted distance distribution and the actual distance distribution is:

[0072]

[0073] The adaptive learning rate adjustment formula is:

[0074]

[0075] in, It is feature maps or feature vectors; It is The weights corresponding to the feature maps or feature vectors are calculated, and these weights can be optimized through the back-propagation algorithm, so that the model can automatically adjust the contribution between different features;

[0076] and The predicted distance and the actual distance fall in The probability within an interval; and They are the predicted distance distribution and the actual distance distribution respectively;

[0077] and are the learning rates of the previous iteration and this iteration respectively, is the learning rate decay coefficient, is a preset positive number. and The loss function is in the parameters and parameters The gradient at .

[0078] This embodiment S1.2 treats each feature as a node in the graph neural network and uses edges to represent the relationship between features for fusion. This method can effectively capture the complex associations and interactions between features. This fusion method is more flexible and powerful than traditional linear combinations or simple splicing, which helps to improve the prediction performance of the model. During the forward propagation of the model, the model is controlled to learn the relative importance of different features through the feature fusion weight learning mechanism, so that the model can automatically adjust the weight of each feature, so as to pay more attention to the key features that have an important impact on the prediction results, and further improve the accuracy and robustness of the model. At the output end of the model, the similarity between the predicted distance distribution and the actual distance distribution is measured to obtain the loss value of the current iteration round. This loss value measurement method can intuitively reflect the prediction performance of the model and provide a reliable basis for subsequent gradient calculations and parameter updates.

[0079] S1.3, based on the sample prediction values ​​of several intermediate models, the query set performance stability index is calculated;

[0080] By calculating the distribution difference of the parameters of different meta-learning tasks in the reproducing kernel Hilbert space, the inter-task knowledge transfer metric is obtained; the calculation principle of the inter-task knowledge transfer metric is that the degree of knowledge transfer between tasks is inversely proportional to their distribution difference in the feature space;

[0081] Introduce a meta-learning weight adjustment factor into the meta-learning task set to dynamically adjust the update amplitude of the parameters according to the difficulty value or importance of each meta-learning task; for example, for tasks with higher difficulty or greater importance, a larger weight adjustment factor can be given to speed up the convergence of the model on these tasks;

[0082] The parameter indicator set is composed of the query set performance stability indicator, the inter-task knowledge transfer metric, and the meta-learning weight adjustment factor;

[0083] The meta-learner is trained based on several intermediate models and parameter indicator sets, and the initial prediction model is output based on the meta-learner after the training.

[0084] Among them, the query set performance stability indicators are:

[0085]

[0086] The measure of knowledge transfer between tasks is:

[0087] ;

[0088]

[0089] The formula for the meta-learning weight adjustment factor is:

[0090]

[0091] in, is the number of samples in the query set, It is Sample prediction values, is the average of the sample prediction values;

[0092] is the measure of knowledge transfer between tasks; It is an intermediate parameter used to measure the difference in the distribution of two data sets, that is, the maximum mean difference; is the source task dataset, is the target task dataset, is the mapping function that maps data to the reproducing kernel Hilbert space, and are the mean vectors of the source task dataset and the target task dataset in the reproducing kernel Hilbert space respectively; is a hyperparameter used to control the extent to which the maximum mean difference affects the knowledge transfer metric between tasks; and Respectively represent the number of samples in the source task dataset and the target task dataset, represents the reproducing kernel Hilbert space.

[0093] represents the update amount of external network parameters, It is The meta-learning weight adjustment factor for each meta-learning task, is the loss function About external network parameters The gradient of

[0094] In this embodiment S1.3, the query set performance stability index can reflect the generalization ability and robustness of the model on different samples. By calculating the distribution difference of the parameters of different meta-learning tasks in the reproducing kernel Hilbert space to obtain the knowledge transfer metric between tasks, the similarities and differences between tasks can be quantified. By introducing the meta-learning weight adjustment mechanism, the update amplitude of the parameters can be dynamically adjusted according to the difficulty value of each meta-learning task.

[0095] In this embodiment S1.1 to S1.3, the internal network model is trained in an inner loop in the meta-learning task set, which helps the model to fully learn the information in the data and improve the generalization ability of the model. The meta-learner can be trained using several intermediate models and a parameter indicator set established according to the query set performance stability index, the inter-task knowledge transfer metric and the meta-learning weight adjustment factor. This method can comprehensively consider information from multiple aspects, so that the meta-learner can learn the characteristics of the data more comprehensively and accurately.

[0096] S1.4, mapping the physical derivative features and original features of the wave parameter data to a high-dimensional space based on a kernel function, and fusing the physical derivative features and original features in the high-dimensional space to obtain a feature set; wherein the physical derivative features of the wave parameter data are obtained by performing frequency domain analysis and statistical analysis on the wave parameter data, for example, statistical features such as wave height distribution and period distribution can be calculated;

[0097] An ensemble learning method is used to train the initial prediction model according to the feature set of wave parameter data to obtain a prediction model; wherein the ensemble learning method includes random forest, gradient boosting tree, etc.

[0098] In this embodiment S1.4, the original features and the physically derived features are fused in a high-dimensional space. This fusion method can make full use of the information of the two types of features to generate a more representative feature set, which helps to improve the prediction accuracy of subsequent models.

[0099] S1.5. Set a time window, calculate the weight value according to the relevant wave data of the wave parameter data in the time window, and continuously update the weight value as the time window slides to obtain the wave weight parameter;

[0100] Fuzzy sets are defined for the two objectives of minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group, and several fuzzy sets including the fuzzy set of the impact effect of waves on the column group and the fuzzy set of the stability coefficient of the column group are obtained; and membership functions are constructed for each of the fuzzy sets in the fuzzy sets, and several membership function sets are obtained; among them, the fuzzy set of the impact effect of waves on the column group is a mapping from the actual impact effect value to the fuzzy set, and the mapping range is from 0 (no impact) to 1 (maximum impact), and the fuzzy set is used to describe the size of the impact effect of waves on the column group, and the smaller the value, the smaller the impact effect; the fuzzy set of the stability coefficient of the column group is a mapping from the actual stability coefficient value to the fuzzy set, and the mapping range is from 0 (lowest stability) to 1 (highest stability), and the fuzzy set is used to describe the level of the stability coefficient of the column group, and the larger the value, the higher the stability;

[0101] Based on wave weight parameters and several membership function sets, a multi-objective optimization function is constructed.

[0102] Among them, "the impact effect of waves on the column group" refers to the dynamic force or pressure generated when the wave contacts and acts on the column group, and the "stability coefficient of the column group" is an indicator used to measure the stability of the column group structure when subjected to external forces; and several membership function sets include the impact effect fuzzy set membership function and the stability coefficient fuzzy set membership function, specifically:

[0103] The membership function of the shock effect fuzzy set is:

[0104]

[0105] The membership function of the stability coefficient fuzzy set is:

[0106]

[0107] in, is the actual maximum value of the shock effect, is the current value of the shock effect; is the actual maximum value of the stability coefficient, is the current value of the stability factor.

[0108] This embodiment S1.5 can capture the dynamic changes of waves in real time by setting a time window and calculating the weight value based on the wave data in the window; and as the time window slides, the weight value is continuously updated, thereby reflecting the latest situation of the wave conditions. Defining fuzzy sets for the two goals of minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group helps to deal with the uncertainty in these goals. A membership function is constructed for each fuzzy set so that the specific value of each goal can be mapped to a membership between 0 and 1. This helps to balance these two potentially conflicting goals during the optimization process.

[0109] S1.6. Input the wave-related data of the target sea area into the prediction model for prediction, and obtain the model prediction result;

[0110] Set the main particle swarm and auxiliary particle swarm according to the model prediction results;

[0111] Based on the multi-objective optimization function, the particle swarm optimization algorithm and the adaptive mechanism are used to search in the main particle swarm to obtain the global optimal solution. The adaptive mechanism refers to dynamically adjusting these parameters according to the historical search performance of the particles and the state of the current search space. For example, when the particle swarm is close to the global optimal solution, the inertia weight is reduced to strengthen the local search; when it falls into the local optimum, the acceleration factor is increased to promote the particles to jump out of the local optimum.

[0112] In the auxiliary particle group, a combination operation is performed on several elite particles and a mutation operation is performed on several elite particles to obtain an auxiliary global optimal solution; wherein several elite particles are obtained by dividing the auxiliary particle group according to the fitness value of the particles;

[0113] The global optimal solution and the auxiliary global optimal solution are substituted into the multi-objective optimization function for calculation, and the objective function values ​​of the global optimal solution and the objective function values ​​of the auxiliary global optimal solution are obtained respectively. These objective function values ​​are compared to evaluate the performance of different solutions in terms of optimization objectives; the global optimal solution or the auxiliary global optimal solution with better objective function value performance is defined as the actual optimal solution, and the actual optimal solution is output as the final model prediction result of the optimal relative arrangement distance of the column group, and a specific construction plan of the column group breakwater is generated according to the final model prediction result of the optimal relative arrangement distance of the column group.

[0114] Use professional structural analysis software (such as SAP2000, ANSYS) to model and analyze the structure under the optimal relative arrangement distance of column groups; evaluate the overall stability and safety of the structure by calculating parameters such as displacement, stress and strain of the structure, so as to obtain the evaluation results of the engineering value; use the preset cost estimation software to estimate the costs of materials, labor, machinery, etc. according to the design plan of the optimal relative arrangement distance of column groups, so as to obtain the evaluation results of economic benefits.

[0115] In this embodiment S1.6, by setting the main particle group and the auxiliary particle group, the search space is effectively divided. The main particle group focuses on global search and can quickly locate the better solution space area. The auxiliary particle group further combines and mutates the elite particles, which helps to perform detailed search in the located solution space area, thereby improving the search efficiency.

[0116] Overall, this embodiment has the following beneficial effects:

[0117] This application can find potential optimal solutions in the entire solution space through the particle swarm optimization algorithm; at the same time, through the local optimization process, the algorithm can continuously approach the optimal solution and improve the quality of the solution. In addition, the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group. This optimization goal directly targets the key performance indicators of the breakwater project, so that the optimization results are more in line with actual needs. For the prediction model, by fusing the physical derivative features and original features of the wave parameter data to construct a feature set, this feature fusion technology can more comprehensively reflect the characteristics and laws of the wave data and improve the prediction accuracy of the model. The inner loop optimizes the model's processing and analysis capabilities for wave data by fusing the weights of several parameter features; this optimization enables the model to more accurately capture subtle changes in wave data and improve the accuracy of prediction. The knowledge transfer metric in the outer loop process is used to further adjust and optimize the model, which helps the model to better utilize historical data and empirical knowledge, improve its applicability in different sea areas, and thus ensure the accuracy of the model prediction;

[0118] Moreover, this application successfully overcomes the challenges of high construction costs and complex technical difficulties faced by the field of breakwater construction by obtaining the final model prediction results of the relative arrangement distance of the optimal column groups, and effectively maintains the balance of the ecological environment in the port area. Furthermore, by using scientific methods to comprehensively evaluate the engineering value and economic benefits, it not only significantly reduces the technical difficulty of construction and shortens the project preparation cycle, but also achieves remarkable results in cost control, while optimizing the functional performance of wave reduction and wave reduction, and achieving a dual improvement in economic and ecological benefits.

[0119] Embodiment 2:

[0120] See also Figure 2 , the embodiment of the present application provides an artificial intelligence-based eco-friendly column group breakwater construction system, including a prediction module 10;

[0121] The prediction module 10 is used to input the wave-related data of the target sea area into the prediction model for prediction, and optimize the model prediction results based on the particle swarm optimization algorithm to obtain the final model prediction results of the relative arrangement distance of the optimal column group; wherein the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of the wave on the column group and maximizing the stability coefficient of the column group;

[0122] Among them, the prediction model is established by learning and training the initial prediction model according to the feature set of wave parameter data; the feature set is established by fusing the physical derivative features of the wave parameter data with the original features; the initial prediction model is based on the wave parameter data of the target sea area and machine learning technology, and is established by fusing the weights of several parameter features through an inner loop, as well as based on the knowledge transfer metric in the outer loop process.

[0123] In one embodiment, the prediction module 10 includes a data unit, a transition unit, a model unit, a feature unit, a function unit, and a prediction unit;

[0124] The data unit is used to regularly record wave data in the target sea area through equipment such as buoys, wave measuring vessels or fixed wave measuring stations, so as to obtain wave parameter data of the target sea area, and clean, normalize or standardize the wave parameter data; wherein the wave parameter data includes key characteristic data such as wave height, wave period, wind direction, wind speed and water depth;

[0125] The data unit is also used to divide the wave parameter data into several subsets according to the time window, so that each subset contains all the wave parameter data in the time window; the divided data subsets are taken as independent tasks to construct a meta-learning task, thereby obtaining a meta-learning task set composed of several meta-learning tasks; wherein the time window can be a fixed length, such as hourly, daily, weekly or monthly, etc.

[0126] A transition unit, used to establish several internal network models based on wave parameter data according to machine learning techniques;

[0127] The transition unit is also used to perform inner loop training on several internal network models in the meta-learning task set, and in the forward propagation process of the model, control the model to learn the relative importance of different features through the feature fusion weight learning mechanism, and perform feature fusion on different features; wherein, feature fusion is performed by treating each feature as a node in the graph neural network and representing the relationship between features through edges, specifically: using the node representation learning ability of the graph neural network, treating each feature as a node in the graph, and representing the relationship between features through edges; then, using the aggregation operation of the graph neural network to update the node representation, thereby obtaining the fused feature representation;

[0128] The transition unit is also used to measure the similarity between the predicted distance distribution and the actual distance distribution at the output of the model to obtain the loss value of the current iteration round;

[0129] The transition unit is also used to calculate the gradient based on the loss value of the current iteration round to obtain the gradient of the loss function relative to the model parameters;

[0130] The transition unit is also used to dynamically adjust the learning rate based on the adaptive learning rate adjustment formula in each iteration of the inner loop training;

[0131] The transition unit is also used to update the parameters of several internal network models according to the gradient and the adjusted learning rate. If the current iteration rounds of the several internal network models reach a preset number of iterations or the loss value converges below a preset threshold, the trained several internal network models are defined as several intermediate models.

[0132] Among them, the formula of the feature fusion weight learning mechanism is:

[0133]

[0134] The formula for measuring the similarity between the predicted distance distribution and the actual distance distribution is:

[0135]

[0136] The adaptive learning rate adjustment formula is:

[0137]

[0138] in, It is feature maps or feature vectors; It is The weights corresponding to the feature maps or feature vectors are calculated, and these weights can be optimized through the back-propagation algorithm, so that the model can automatically adjust the contribution between different features;

[0139] and The predicted distance and the actual distance fall in The probability within an interval; and They are the predicted distance distribution and the actual distance distribution respectively;

[0140] and are the learning rates of the previous iteration and this iteration respectively, is the learning rate decay coefficient, is a preset positive number. and The loss function is in the parameters and parameters The gradient at .

[0141] The transition unit of this embodiment treats each feature as a node in the graph neural network and uses edges to represent the relationship between features for fusion. This method can effectively capture the complex associations and interactions between features. This fusion method is more flexible and powerful than traditional linear combinations or simple splicing, and helps to improve the prediction performance of the model. During the forward propagation of the model, the model is controlled to learn the relative importance of different features through the feature fusion weight learning mechanism, so that the model can automatically adjust the weight of each feature, so as to pay more attention to the key features that have an important impact on the prediction results, and further improve the accuracy and robustness of the model. At the output end of the model, the similarity between the predicted distance distribution and the actual distance distribution is measured to obtain the loss value of the current iteration round. This loss value measurement method can intuitively reflect the prediction performance of the model and provide a reliable basis for subsequent gradient calculations and parameter updates.

[0142] A model unit is used to calculate the query set performance stability index based on the sample prediction values ​​of several intermediate models;

[0143] The model unit is also used to obtain the inter-task knowledge transfer metric by calculating the distribution difference of the parameters of different meta-learning tasks in the reproducing kernel Hilbert space; wherein the calculation principle of the inter-task knowledge transfer metric is that the degree of knowledge transfer between tasks is inversely proportional to their distribution difference in the feature space;

[0144] The model unit is also used to introduce a meta-learning weight adjustment factor into the meta-learning task set to dynamically adjust the update amplitude of the parameters according to the difficulty value or importance of each meta-learning task; for example, for tasks with higher difficulty or greater importance, a larger weight adjustment factor can be given to speed up the convergence of the model on these tasks;

[0145] The model unit is also used to form a parameter indicator set consisting of a query set performance stability indicator, an inter-task knowledge transfer metric, and a meta-learning weight adjustment factor;

[0146] The model unit is also used to train the meta-learner based on several intermediate models and parameter indicator sets, and output an initial prediction model according to the meta-learner after the training.

[0147] Among them, the query set performance stability indicators are:

[0148]

[0149] The measure of knowledge transfer between tasks is:

[0150] ;

[0151]

[0152] The formula for the meta-learning weight adjustment factor is:

[0153]

[0154] in, is the number of samples in the query set, It is Sample prediction values, is the average of the sample prediction values;

[0155] is the measure of knowledge transfer between tasks; It is an intermediate parameter used to measure the difference in the distribution of two data sets, that is, the maximum mean difference; is the source task dataset, is the target task dataset, is the mapping function that maps data to the reproducing kernel Hilbert space, and are the mean vectors of the source task dataset and the target task dataset in the reproducing kernel Hilbert space respectively; is a hyperparameter used to control the extent to which the maximum mean difference affects the knowledge transfer metric between tasks; and Respectively represent the number of samples in the source task dataset and the target task dataset, represents the reproducing kernel Hilbert space.

[0156] represents the update amount of external network parameters, It is The meta-learning weight adjustment factor for each meta-learning task, is the loss function About external network parameters The gradient of

[0157] In the model unit of this embodiment, the query set performance stability index can reflect the generalization ability and robustness of the model on different samples. By calculating the distribution difference of the parameters of different meta-learning tasks in the reproducing kernel Hilbert space to obtain the knowledge transfer metric between tasks, the similarities and differences between tasks can be quantified. By introducing the meta-learning weight adjustment mechanism, the update amplitude of the parameters can be dynamically adjusted according to the difficulty value of each meta-learning task.

[0158] In this embodiment, the data unit, transition unit and model unit perform inner loop training on the internal network model in the meta-learning task set, which helps the model to fully learn the information in the data and improve the generalization ability of the model. The meta-learner can be trained by using several intermediate models and a parameter indicator set established according to the query set performance stability index, the inter-task knowledge transfer metric and the meta-learning weight adjustment factor. This method can comprehensively consider information from multiple aspects, so that the meta-learner can learn the characteristics of the data more comprehensively and accurately.

[0159] A feature unit is used to map the physical derivative features and original features of the wave parameter data to a high-dimensional space based on a kernel function, and to fuse the physical derivative features and the original features in the high-dimensional space to obtain a feature set; wherein the physical derivative features of the wave parameter data are obtained by performing frequency domain analysis and statistical analysis on the wave parameter data, for example, statistical features such as wave height distribution and period distribution can be calculated;

[0160] The feature unit is also used to use an integrated learning method to learn and train the initial prediction model according to the feature set of the wave parameter data to obtain a prediction model; wherein the integrated learning method includes random forest, gradient boosting tree, etc.

[0161] The feature unit of this embodiment fuses the original features and the physically derived features in a high-dimensional space. This fusion method can fully utilize the information of the two types of features to generate a more representative feature set, which helps to improve the prediction accuracy of subsequent models.

[0162] The function unit is used to set a time window, calculate the weight value according to the relevant wave data of the wave parameter data in the time window, and continuously update the weight value as the time window slides to obtain the wave weight parameter;

[0163] The function unit is also used to define fuzzy sets for the two objectives of minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group, and obtain several fuzzy sets including the fuzzy set of the impact effect of waves on the column group and the fuzzy set of the stability coefficient of the column group; and construct a membership function for each fuzzy set in the several fuzzy sets, and obtain several membership function sets; wherein the fuzzy set of the impact effect of waves on the column group is a mapping from the actual impact effect value to the fuzzy set, and the mapping range is from 0 (no impact) to 1 (maximum impact), and the fuzzy set is used to describe the size of the impact effect of waves on the column group, and the smaller the value, the smaller the impact effect; the fuzzy set of the stability coefficient of the column group is a mapping from the actual stability coefficient value to the fuzzy set, and the mapping range is from 0 (lowest stability) to 1 (highest stability), and the fuzzy set is used to describe the level of the stability coefficient of the column group, and the larger the value, the higher the stability;

[0164] The function unit is also used to construct a multi-objective optimization function based on wave weight parameters and several membership function sets.

[0165] Among them, "the impact effect of waves on the column group" refers to the dynamic force or pressure generated when the wave contacts and acts on the column group, and the "stability coefficient of the column group" is an indicator used to measure the stability of the column group structure when subjected to external forces; and several membership function sets include the impact effect fuzzy set membership function and the stability coefficient fuzzy set membership function, specifically:

[0166] The membership function of the shock effect fuzzy set is:

[0167]

[0168] The membership function of the stability coefficient fuzzy set is:

[0169]

[0170] in, is the actual maximum value of the shock effect, is the current value of the shock effect; is the actual maximum value of the stability coefficient, is the current value of the stability factor.

[0171] The function unit of this embodiment can capture the dynamic changes of waves in real time by setting a time window and calculating the weight value according to the wave data in the window; and as the time window slides, the weight value is continuously updated, thereby reflecting the latest situation of the wave conditions. Defining fuzzy sets for the two goals of minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group helps to deal with the uncertainty in these goals. A membership function is constructed for each fuzzy set so that the specific value of each goal can be mapped to a membership between 0 and 1. This helps to balance these two potentially conflicting goals during the optimization process.

[0172] A prediction unit is used to input wave-related data of the target sea area into a prediction model for prediction and obtain a model prediction result;

[0173] The prediction unit is also used to set the main particle swarm and the auxiliary particle swarm according to the model prediction results;

[0174] The prediction unit is also used to search in the main particle swarm based on the multi-objective optimization function according to the particle swarm optimization algorithm and the adaptive mechanism to obtain the global optimal solution; the adaptive mechanism refers to dynamically adjusting these parameters according to the historical search performance of the particles and the state of the current search space. For example, when the particle swarm is close to the global optimal solution, the inertia weight is reduced to strengthen the local search; when it falls into the local optimum, the acceleration factor is increased to promote the particles to jump out of the local optimum;

[0175] The prediction unit is also used to perform a combination operation on a number of elite particles and a mutation operation on a number of elite particles in the auxiliary particle group to obtain an auxiliary global optimal solution; wherein the number of elite particles are obtained by dividing the auxiliary particle group according to the fitness value of the particles;

[0176] The prediction unit is also used to substitute the global optimal solution and the auxiliary global optimal solution into the multi-objective optimization function for calculation, and obtain the objective function value of the global optimal solution and the objective function value of the auxiliary global optimal solution respectively, and compare these objective function values ​​to evaluate the performance of different solutions on the optimization objectives; define the global optimal solution or the auxiliary global optimal solution with better objective function value performance as the actual optimal solution, and output the actual optimal solution as the final model prediction result of the relative arrangement distance of the optimal column group, and generate a specific construction plan of the column group breakwater according to the final model prediction result of the relative arrangement distance of the optimal column group.

[0177] The prediction unit is also used to use professional structural analysis software (such as SAP2000, ANSYS) to model and analyze the structure under the optimal relative arrangement distance of column groups; by calculating the displacement, stress, strain and other parameters of the structure, the overall stability and safety of the structure are evaluated to obtain the evaluation results of the engineering value; the preset cost estimation software is used to estimate the costs of materials, labor, machinery, etc. according to the design plan of the optimal relative arrangement distance of column groups, so as to obtain the evaluation results of economic benefits.

[0178] The prediction unit of this embodiment realizes the effective division of the search space by setting the main particle group and the auxiliary particle group. The main particle group focuses on global search and can quickly locate the better solution space area. The auxiliary particle group further combines and mutates the elite particles, which helps to perform fine search in the located solution space area, thereby improving the search efficiency.

[0179] Overall, this embodiment has the following beneficial effects:

[0180] This application can find potential optimal solutions in the entire solution space through the particle swarm optimization algorithm; at the same time, through the local optimization process, the algorithm can continuously approach the optimal solution and improve the quality of the solution. In addition, the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group. This optimization goal directly targets the key performance indicators of the breakwater project, so that the optimization results are more in line with actual needs. For the prediction model, by fusing the physical derivative features and original features of the wave parameter data to construct a feature set, this feature fusion technology can more comprehensively reflect the characteristics and laws of the wave data and improve the prediction accuracy of the model. The inner loop optimizes the model's processing and analysis capabilities for wave data by fusing the weights of several parameter features; this optimization enables the model to more accurately capture subtle changes in wave data and improve the accuracy of prediction. The knowledge transfer metric in the outer loop process is used to further adjust and optimize the model, which helps the model to better utilize historical data and empirical knowledge, improve its applicability in different sea areas, and thus ensure the accuracy of the model prediction;

[0181] Moreover, this application successfully overcomes the challenges of high construction costs and complex technical difficulties faced by the field of breakwater construction by obtaining the final model prediction results of the relative arrangement distance of the optimal column groups, and effectively maintains the balance of the ecological environment in the port area. Furthermore, by using scientific methods to comprehensively evaluate the engineering value and economic benefits, it not only significantly reduces the technical difficulty of construction and shortens the project preparation cycle, but also achieves remarkable results in cost control, while optimizing the functional performance of wave reduction and wave reduction, and achieving a dual improvement in economic and ecological benefits.

[0182] Embodiment three:

[0183] The embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for constructing an eco-friendly column group breakwater based on artificial intelligence;

[0184] Wherein, the method for constructing an eco-friendly column group breakwater based on artificial intelligence, if implemented in the form of a software functional unit and used as an independent product, can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0185] The above are preferred embodiments of the present invention. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based eco-friendly column-group breakwater construction method, characterized in that: include: The wave-related data of the target sea area is input into the prediction model for prediction, and the model prediction results are optimized based on the particle swarm optimization algorithm to obtain the final model prediction results of the optimal relative arrangement distance of the column group; wherein the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of the wave on the column group and maximizing the stability coefficient of the column group; The prediction model is established by training the initial prediction model based on the feature set of wave parameter data; the feature set is established by fusing the physical derivative features of the wave parameter data with the original features; the initial prediction model is based on the wave parameter data of the target sea area and machine learning technology, by fusing the weights of several parameter features through an inner loop, and by establishing the knowledge transfer metric value in the outer loop process; The particle swarm optimization algorithm is used to optimize the model prediction results to obtain the final model prediction results of the optimal column group relative arrangement distance, which are specifically: Setting a main particle group and an auxiliary particle group according to the prediction results of the model; Searching in the master particle swarm according to the particle swarm optimization algorithm to obtain a global optimal solution; In the auxiliary particle group, a combination operation is performed on a number of elite particles and a mutation operation is performed on the number of elite particles to obtain an auxiliary global optimal solution; wherein the number of elite particles are obtained by dividing the auxiliary particle group according to the fitness values ​​of the particles; The actual optimal solution among the global optimal solution and the auxiliary global optimal solution is defined as the final model prediction result regarding the relative arrangement distance of the optimal column groups.

2. The method for constructing an eco-friendly column-group breakwater based on artificial intelligence as claimed in claim 1, characterized in that: The initial prediction model is based on the wave parameter data of the target sea area and machine learning technology, and is established by fusing the weights of several parameter features through an inner loop and based on the knowledge transfer metric value in the outer loop process, specifically: Acquiring the wave parameter data of the target sea area; Dividing the wave parameter data into a plurality of subsets according to a time window, and obtaining a meta-learning task set consisting of the plurality of subsets; In the meta-learning task set, several internal network models are trained in an inner loop, and several parameter feature weights in the model forward propagation process are fused to obtain several intermediate models; wherein the several internal network models are established based on the wave parameter data according to machine learning technology; The meta-learner is trained based on the several intermediate models and parameter indicator sets, and the initial prediction model is outputted by the meta-learner after the training; wherein the parameter indicator set is established based on the query set performance stability index, the inter-task knowledge transfer metric and the meta-learning weight adjustment factor.

3. The method for constructing an eco-friendly column-group breakwater based on artificial intelligence as claimed in claim 2, characterized in that: Several internal network models are trained in an inner loop, and several parameter feature weights in the forward propagation process of the model are fused to obtain several intermediate models, specifically: The plurality of internal network models are trained in an inner loop, and in the forward propagation process of the model, the model is controlled to learn the relative importance of different features through a feature fusion weight learning mechanism, and different features are fused; wherein the feature fusion is performed by treating each feature as a node in a graph neural network and representing the relationship between the features through edges; At the output of the model, measure the similarity between the predicted distance distribution and the actual distance distribution to obtain the loss value of the current iteration round; Performing gradient calculation based on the loss value of the current iteration round to obtain the gradient of the loss function relative to the model parameters; In each iteration of the inner loop training, dynamically adjusting the learning rate based on an adaptive learning rate adjustment formula; Parameters of the several internal network models are updated according to the gradient and the adjusted learning rate. If the current iteration rounds of the several internal network models reach a preset number of iterations or the loss value converges below a preset threshold, the several internal network models after training are defined as the several intermediate models.

4. The method for constructing an eco-friendly column-group breakwater based on artificial intelligence as claimed in claim 3, characterized in that: The adaptive learning rate adjustment formula is specifically: ； in, and are the learning rates of the previous iteration and this iteration respectively, is the learning rate decay coefficient, is a preset positive number. and The loss function is in the parameters and parameters The gradient at .

5. The method for constructing an eco-friendly column-group breakwater based on artificial intelligence as claimed in claim 2, characterized in that: The parameter indicator set is established based on the query set performance stability indicator, the inter-task knowledge transfer metric and the meta-learning weight adjustment factor, specifically: Based on the sample prediction values ​​of the several intermediate models, the query set performance stability index is calculated; By calculating the distribution difference of parameters of different meta-learning tasks in the reproducing kernel Hilbert space, a knowledge transfer metric value between the tasks is obtained; According to the difficulty value of each meta-learning task, dynamically adjust the update amplitude of the parameters by introducing the meta-learning weight adjustment factor; The parameter indicator set is composed of the query set performance stability indicator, the inter-task knowledge transfer metric and the meta-learning weight adjustment factor.

6. The method for constructing an eco-friendly column-group breakwater based on artificial intelligence as claimed in claim 5, characterized in that: The inter-task knowledge transfer metric is specifically: ； ； in, is the knowledge transfer metric between the tasks, is an intermediate parameter used to measure the difference in the distribution of two data sets. is the source task dataset, is the target task dataset, is the mapping function that maps data to the reproducing kernel Hilbert space, and are the mean vectors of the source task dataset and the target task dataset in the reproducing kernel Hilbert space, is a hyperparameter, and Respectively represent the number of samples in the source task dataset and the target task dataset, represents the reproducing kernel Hilbert space.

7. The method for constructing an eco-friendly column-group breakwater based on artificial intelligence as claimed in claim 1, characterized in that: The multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group, specifically: A time window is set, a weight value is calculated according to the wave data in the time window, and the weight value is continuously updated as the time window slides to obtain a wave weight parameter; Fuzzy sets are defined for the two objectives of minimizing the impact effect of waves on the column group and maximizing the stability coefficient of the column group, and a membership function is constructed for each fuzzy set to obtain a membership function set; Based on the wave weight parameters and the membership function set, the multi-objective optimization function is constructed.

8. The method for constructing an eco-friendly column-group breakwater based on artificial intelligence as claimed in claim 1, characterized in that: The feature set is established by fusing the physically derived features of the wave parameter data with the original features, specifically: The physical derivative features and the original features of the wave parameter data are mapped to a high-dimensional space based on a kernel function, and the physical derivative features and the original features are fused in the high-dimensional space to obtain the feature set.

9. An artificial intelligence-based eco-friendly column breakwater construction system, characterized in that: Includes prediction module; The prediction module is used to input the wave-related data of the target sea area into the prediction model for prediction, and optimize the model prediction results based on the particle swarm optimization algorithm to obtain the final model prediction results of the relative arrangement distance of the optimal column group; wherein the multi-objective optimization function of the particle swarm optimization algorithm is established by minimizing the impact effect of the wave on the column group and maximizing the stability coefficient of the column group; The prediction model is established by training the initial prediction model based on the feature set of wave parameter data; the feature set is established by fusing the physical derivative features of the wave parameter data with the original features; the initial prediction model is based on the wave parameter data of the target sea area and machine learning technology, by fusing the weights of several parameter features through an inner loop, and by establishing the knowledge transfer metric value in the outer loop process; The particle swarm optimization algorithm is used to optimize the model prediction results to obtain the final model prediction results of the optimal column group relative arrangement distance, which are specifically: Setting a main particle group and an auxiliary particle group according to the prediction results of the model; Searching in the master particle swarm according to the particle swarm optimization algorithm to obtain a global optimal solution; In the auxiliary particle group, a combination operation is performed on a number of elite particles and a mutation operation is performed on the number of elite particles to obtain an auxiliary global optimal solution; wherein the number of elite particles are obtained by dividing the auxiliary particle group according to the fitness values ​​of the particles; The actual optimal solution among the global optimal solution and the auxiliary global optimal solution is defined as the final model prediction result regarding the relative arrangement distance of the optimal column groups.

Images